Subspace Change-Point Detection: A New Model and Solution

Wikipedia has been turned into an immensely popular crowd-sourced encyclopedia for information dissemination on numerous versatile topics in the form of subscription free content. It allows anyone to contribute so that the articles remain comprehensive and updated. For enrichment of content without compromising standards, the Wikipedia community enumerates a detailed set of guidelines, which should be followed. Based on these, articles are categorized into several quality classes by the Wikipedia editors with increasing adherence to guidelines. This quality assessment task by editors is laborious as well as demands platform expertise. As a first objective, in this paper, we study evolution of a Wikipedia article with respect to such quality scales. Our results show novel non-intuitive patterns emerging from this exploration. As a second objective we attempt to develop an automated data driven approach for the detection of the early signals influencing the quality change of articles. We posit this as a change point detection problem whereby we represent an article as a time series of consecutive revisions and encode every revision by a set of intuitive features. Finally, various change point detection algorithms are used to efficiently and accurately detect the future change points. We also perform various ablation studies to understand which group of features are most effective in identifying the change points. To the best of our knowledge, this is the first work that rigorously explores English Wikipedia article quality life cycle from the perspective of quality indicators and provides a novel unsupervised page level approach to detect quality switch, which can help in automatic content monitoring in Wikipedia thus contributing significantly to the CSCW community.

Many modern applications require detecting change points in complex sequential data. Most existing methods for change point detection used in streaming data settings are unsupervised and, as a consequence, lack any information regarding what kind of changes we want to detect or if some kinds of changes are safe to ignore. This often results in poor change detection performance, as many spurious changes are falsely detected as change points. We present a novel change point detection framework that uses a small number of labeled change point instances as supervision for learning a ground metric such that Sinkhorn divergences can be then used in two-sample tests on sliding windows to detect change points. Our method can be used to learn a sparse metric which can be useful for both feature selection and interpretation in high-dimensional change point detection settings. Experiments on simulated as well as real world sequences show that our proposed method can substantially improve change point detection performance over existing unsupervised change point detection methods using only few labeled change point instances.

Sliding window change point detection based dynamic network model inference framework for airport ground service process

Early detection of fuel leakage at service stations with underground petroleum storage systems is a crucial task to prevent catastrophic hazards. Current data-driven fuel leakage detection methods employ offline statistical inventory reconciliation, leading to significant detection delays. Consequently, this can result in substantial financial loss and environmental impact on the surrounding community. In this paper, we propose a novel framework called memory-based online change point detection (MOCPD) which operates in near real-time, enabling early detection of fuel leakage. MOCPD maintains a collection of representative historical data within a size-constrained memory, along with an adaptively computed threshold. Leaks are detected when the dissimilarity between the latest data and historical memory exceeds the current threshold. An update phase is incorporated in MOCPD to ensure diversity among historical samples in the memory. With this design, MOCPD is more robust and achieves a better recall rate while maintaining a reasonable precision score. We have conducted a variety of experiments comparing MOCPD to commonly used online change point detection (CPD) baselines on real-world fuel variance data with induced leakages, actual fuel leakage data and benchmark CPD datasets. Overall, MOCPD consistently outperforms the baseline methods in terms of detection accuracy, demonstrating its applicability to fuel leakage detection and CPD problems.

Change Point Detection (CPD) aims to identify moments of abrupt distribution shifts in data streams. Real-world high-dimensional CPD remains challenging due to data pattern complexity and violation of common assumptions. Resorting to standalone deep neural networks, the current state-of-theart detectors have yet to achieve perfect quality. Concurrently, ensembling provides more robust solutions, boosting the performance. In this paper, we investigate ensembles of deep change point detectors and realize that standard prediction aggregation techniques, e.g., averaging, are suboptimal and fail to account for problem peculiarities. Alternatively, we introduce WWAggr - a novel task-specific method of ensemble aggregation based on the Wasserstein distance. Our procedure is versatile, working effectively with various ensembles of deep CPD models. Moreover, unlike existing solutions, we practically lift a long-standing problem of the decision threshold selection for CPD.

Modern industrial assets (e.g., generators, turbines, engines) are outfitted with numerous sensors to monitor key operating and environmental variables. Unusual sensor readings, such as high temperature, excessive vibration, or low current, could trigger rule-based actions (also known as faults) that range from warning alarms to immediate shutdown of the asset to prevent potential damage. In the case study of this article, a wind park experienced a sudden surge in vibration-induced shutdowns. We utilize fault data logs from the park with the goal of detecting common change points across turbines. Another important goal is the localization of fault occurrences to an identifiable set of turbines. The literature on change point detection and localization for multiple assets is highly sparse. Our technical development is based on the generalized linear modeling framework. We combine well-known solutions to change point detection for a single asset with a heuristics-based approach to identify a common change point(s) for multiple assets. The performance of the proposed detection and localization algorithms is evaluated through synthetic (Monte Carlo) fault data streams. Several novel performance metrics are defined to characterize different aspects of a change point detection algorithm for multiple assets. For the case study example, the proposed methodology identified the change point and the subset of affected turbines with a high degree of accuracy. The problem described here warrants further study to accommodate general fault distributions, change point detection algorithms, and very large fleet sizes.

Economic system mathematical model often contains multiple variance change points about structure model. In the same mean, we combine the Bayesian method with the maximum likelihood method on the detection of the variance multiple change points. With Bayesian method, we can eliminate extra parameters first, and then use maximum likelihood method to find the change position. So we both can eliminate extra parameters and can avoid the change point on the prior distribution unknown problem. In addition, the benefit of the maximum likelihood method is just needed to find out likelihood density function of the maximal solution in the solution space, thus the variance of multiple change point detection problem is resolved. By making substantial analysis with an example from commodity house prices in Wuhan, three changes of variance are found in all. They correspond to the major structure changes of contemporary property market in Wuhan city. The method is practical and effective as the final results shown. It also has a certain practical significance.

MotivationConventional identification methods for gene regulatory networks (GRNs) have overwhelmingly adopted static topology models, which remains unchanged over time to represent the underlying molecular interactions of a biological system. However, GRNs are dynamic in response to physiological and environmental changes. Although there is a rich literature in modeling static or temporally invariant networks, how to systematically recover these temporally changing networks remains a major and significant pressing challenge. The purpose of this study is to suggest a two-step strategy that recovers time-varying GRNs.ResultsIt is suggested in this paper to utilize a switching auto-regressive model to describe the dynamics of time-varying GRNs, and a two-step strategy is proposed to recover the structure of time-varying GRNs. In the first step, the change points are detected by a Kalman-filter based method. The observed time series are divided into several segments using these detection results; and each time series segment belonging to two successive demarcating change points is associated with an individual static regulatory network. In the second step, conditional network structure identification methods are used to reconstruct the topology for each time interval. This two-step strategy efficiently decouples the change point detection problem and the topology inference problem. Simulation results show that the proposed strategy can detect the change points precisely and recover each individual topology structure effectively. Moreover, computation results with the developmental data of Drosophila Melanogaster show that the proposed change point detection procedure is also able to work effectively in real world applications and the change point estimation accuracy exceeds other existing approaches, which means the suggested strategy may also be helpful in solving actual GRN reconstruction problem.

Capturing correlation changes by applying kernel change point detection on the running correlations

Environmental health studies are of great interest in research to evaluate the mortality-temperature relationship by adjusting spatially correlated random effects as well as identifying significant change points in temperature. However, this relationship is often not expressed using parametric models, which makes identifying change points an even more challenging problem. This paper proposes a unified semiparametric approach to simultaneously identify the nonlinear mortality-temperature relationship and detect spatially-dependent change points. A unified method is proposed for the model estimation, spatially dependent change points detection, and testing whether they are significant simultaneously by a permutation-based test. We operate under the assumption that change points remain constant, yet acknowledge the uncertainty regarding their precise number. These change points are influenced by the smoothing of an unknown function, which in turn relies on a smoothing variable and spatial random effects. Consequently, the detection of change points may be influenced by spatial effects. In this paper, several simulation studies are conducted to evaluate the performance of our proposed approach. The advantages of this unified approach are demonstrated using epidemiological data on mortality and temperature.

Change point detection for high-dimensional data is an important yet challenging problem for many applications. In this paper, we consider multiple change point detection in the context of high-dimensional generalized linear models, allowing the covariate dimension p to grow exponentially with the sample size n. The model considered is general and flexible in the sense that it covers various specific models as special cases. It can automatically account for the underlying data generation mechanism without specifying any prior knowledge about the number of change points. Based on dynamic programming and binary segmentation techniques, two algorithms are proposed to detect multiple change points, allowing the number of change points to grow with n. To further improve the computational efficiency, a more efficient algorithm designed for the case of a single change point is proposed. We present theoretical properties of our proposed algorithms, including estimation consistency for the number and locations of change points as well as consistency and asymptotic distributions for the underlying regression coefficients. Finally, extensive simulation studies and application to the Alzheimer's Disease Neuroimaging Initiative data further demonstrate the competitive performance of our proposed methods.

The subject of this paper is the detection of a change in the marginal distribution of a stationary linear process. By considering the marginal distribution, the change-point model can simultaneously incorporate any change in the coefficients and/or the innovations of the linear process. Furthermore, the change point can be random and data dependent. The key is an analysis of the asymptotic behaviour of the sequential empirical process, both with and without a change point. Our results hold under very mild conditions on the existence of any moment of the innovations and a corresponding condition of summability of the coefficients.

This paper addresses the change point detection problem in time series. A methodology based on the Metropolis-Hastings algorithm applied to time series modeled as a process with Beta distribution is discussed. In order to make this methodology useful in practice, a fuzzy cluster technique is applied to the initial time series at first, generating a new data set with Beta distribution. Bayesian procedures are considered for inference and the Metropolis-Hastings algorithm is used to sample from the posteriors. In the clustering process, a Kohonen neural network is used having as objective to find the best centers of the time series to be used in the fuzzyfication process. Finally, it will be presented a simulation results in the series of the electric energy consumption in Brazil, between January of 1976 and December of 2000, five months before the blackout occurred in 2001. Such result illustrates the efficiency of the proposed methodology for change point detection in time series.

As video content is responsible for more than 70% of the global IP traffic, related resource allocation approaches, e.g., using content caching, become increasingly important. In this context, to avoid under-provisioning, it is important to rapidly detect and respond to changes in content popularity dynamics, including volatility, i.e., changes in the second order moment of the underlying process. In this paper, we focus on the early identification of changes in the variance of video content popularity, which we address as a statistical change point (CP) detection problem. Unlike changes in the mean that can be well captured by non-parametric statistical approaches, to address this more demanding problem, we construct a hypothesis test that uses in the test statistic both parametric and non-parametric approaches. In the context of parametric models, we consider linear, in the form of autoregressive moving average (ARMA), and, nonlinear, in the form of generalized autoregressive conditional heteroskedasticity (GARCH) processes. We propose an integrated algorithm that combines off-line and on-line CP schemes, with the off-line scheme used as a training (learning) phase. The algorithm is first assessed over synthetic data; our analysis demonstrates that non parametric and GARCH model based approaches can better generalize and are better suited for content views time series with unknown statistics. Finally, the non-parametric and the GARCH based variations of our proposed integrated algorithm are applied on real YouTube video content views time series, to illustrate the performance of the proposed approach of volatility change detection.

In this dissertation a new model for the description of cell organelle movement is defined and a test and change point detection algorithm for changes in the model parameters is proposed. The description of movement patterns can be important for the understanding of various biological processes on multiple scales. One of the primary goals is to understand the causes of change between movement patterns. On the micro scale, movement patterns of cell organelles and swimming micro-organisms such as cells are investigated. To learn more about the different movement types of cell organelles, as well as the changes between these types, we analyse the movement of two specific types of cell organelles in the root of the plant Arabidopsis thaliana, Plastids and Peroxisomes. Interestingly, while all tracks were recorded in three dimensions, over 90% of the tracks show more than 95% of their movement variability in only two dimensions. We therefore focus on the two-dimensional projection of the movements, allowing comparability to approaches for animal movement pattern analysis. In this data set of organelle movement we observe visually prominent, linear movement structures with seemingly piecewise constant movement direction and speed. Since the different sections of the movement could be associated with different movement mechanisms such as movement through cytoplasmic streaming or transport along intracellular filaments, it is of interest to detect the change points (CPs) in the direction and speed of the movement. The most widely used time discrete models are variants of the random walk. Most often used, particularly in animal studies, are so-called correlated random walks, which are parametrised by a random turning angle relative to the previous direction. Correlated random walks are useful to model movement where the difference between highly directed and undirected movement is of interest. By fitting HMM models to the observed organelle movement we show that a biased random walk (BRW), parametrised via absolute directions, may be more appropriate to model sections of different but constant movement direction. The BRW is therefore used as a reference model. However, our findings indicate that a BRW shows a higher variability in the movement direction and may thus move less strictly along a linear structure than can be observed in many organelle tracks. Therefore, we define a new model termed linear walk (LW) with less variability around an expected position. Both models, the BRW and LW, have the same process of expected positions, which is parametrised by two parameters, the movement direction and the step length, where the direction is the angle relative to the x-axis. The models differ in regard to the variability around the expected position. In the BRW, independent random increments are summed up, while in the LW an independent, random error is added to each expected position. Due to this definition the changes in movement direction and step length can be described independently. The expectation of the increments in both models can be parameterised by the movement direction and step length or alternatively by a two dimensional expectation. The maximum likelihood estimators of the model parameters in both models and proof of their strong consistency is provided. Note that for the expectation of the increments in the BRW the estimator is the classical mean, while in the LW the estimator is a weighted average. In the context of the BRW model a known MOSUM approach for the bivariate detection of change points is easily adapted to our setting. In this approach a double window is shifted over the movement track. The difference of the bivariate expectation of increments in the left and right window half is estimated, leading to a process of differences that fluctuates around the origin, but shows systematic deviation in the neighborhood around a CP. Therefore the maximum deviation of the process of differences from the origin is used as a test statistic. The rejection threshold is obtained via simulation of a limit process. CPs are estimated successively by identifying local deviations of the process of differences from the origin. In the LW model the MOSUM has inconvenient properties due to the dependency structure of the increments. Therefore a moving kernel approach is proposed where the maximum likelihood estimators for the expectation in the LW model are used instead of the classical mean. For this approach the proof of the weak convergence of the process of differences assuming the true variance is provided. For both approaches simulation studies concerning the test power and precision are provided. Since the CPs in both approaches are detected within the expectation, we propose a graphical technique to classify the detected change points into CPs in movement direction and step length and apply this technique to the observed organelle movement.